DeepSeek V4 Pro pricing on 59API: pay-per-token vs cards
Your AI bill should follow your code, not the calendar
Most developers meet AI coding tools through a subscription: a fixed monthly price, a bundled allowance, and a reset date on the calendar. That structure is easy to understand. It also sits badly with real coding work, where demand arrives in bursts. A quiet afternoon of refactoring, a week of waiting on a slow CI job, then an afternoon where you are rewriting a service from three different angles.
Consider a hypothetical solo developer. In week one they ask a model to explain a single confusing function. In week two they push a feature through a dozen files. A subscription charges the same for both weeks, so the quiet week subsidises the busy one. You either keep paying for capacity you did not use, or you ration a tool you need most when the workload spikes.
Pay-per-token flips that logic. There is no fixed fee for the month, no allowance to exhaust, and no penalty for a week when usage is light. You pay for the tokens your requests actually consume. For uneven coding workloads, that means cost tracks intensity instead of the billing cycle.
That is the core reason to look at 59API, an independent third-party AI API relay. It gives one account, one API key and one API host for Claude, GPT, Gemini, DeepSeek and other models, with a prepaid balance charged per request by real token usage. No monthly fee. Your balance never expires.
The rest of this guide looks at what that means in practice: how per-token rates work, how DeepSeek V4 Pro pricing compares with other options, and when an optional card makes more sense than pay-as-you-go.
What pay-as-you-go actually looks like on 59API
Per-token billing sounds technical, but the operating model is simple. 59API keeps a prepaid balance on your account, and each request is charged according to the real tokens it uses. There is no monthly fee and no subscription allowance to track. If you make five calls, you pay for five calls. If you make five hundred, you pay for five hundred.
Rates are quoted per 1 million tokens and split into input, output and cache read. Billing is in CNY, with USD figures shown at ¥7.2 = $1. That makes it easy to estimate a request before you send it and to see where a long session is spending.
Top-ups are deliberately small enough to test and large enough to stay out of your way: ¥10, ¥50, ¥100, ¥200, ¥500, ¥1000. You can pay by WeChat Pay or bank card, and invoices are available for teams that need them.
For a developer with an unpredictable week, this structure removes two forms of waste. There is no fee for the months you barely use the API, and there is no stranded subscription when a project ends. Unused balance stays on the account instead of expiring.
One account and one API key also mean you are not juggling credentials across several providers while you decide which model suits the task. The same host, https://api.59api.com, fronts the model catalogue, so moving from a quick model to a stronger one is a configuration change rather than a new billing relationship.
Cards exist too, and we will cover them shortly. But the default 59API path is deliberately unglamorous: top up, make a request, pay for what you used.
DeepSeek V4 Pro pricing and lighter options for lighter tasks
Here is where uneven workloads become visible in the numbers. DeepSeek V4 Pro is priced at $1.25 (¥9) per 1 million input tokens, $3.75 (¥27) per 1 million output tokens and $0.0417 (¥0.3) for a cache read. DeepSeek V4 Flash is priced lower, at $0.2778 (¥2) input, $1.11 (¥8) output and $0.006944 (¥0.05) cache read, which makes it a natural fit for routine edits, quick explanations and repetitive chores.
- DeepSeek V4 Pro: $1.25 (¥9) input, $3.75 (¥27) output per 1 million tokens
- DeepSeek V4 Flash: $0.2778 (¥2) input, $1.11 (¥8) output per 1 million tokens
- Qwen3.7 Plus: $0.1389 (¥1) input, $0.4167 (¥3) output per 1 million tokens
At the lighter end, Qwen3.7 Plus is $0.1389 (¥1) input, $0.4167 (¥3) output and $0.0417 (¥0.3) cache read. MiniMax M3 sits at $0.1389 (¥1) input, $0.5556 (¥4) output and $0.0278 (¥0.2) cache read. When a quiet day only needs a second opinion, these rates keep the meter slow.
The same logic applies to heavier sessions. Take GPT-6.1 Sol: 5M input tokens plus 1M output tokens without cache cost $1.85 (¥13.32) on 59API. At the official list price of $20 (about ¥144), that same workload is about 9% of list price. Official list prices exist only for Claude and GPT, so treat them as reference prices for those families rather than a baseline for every model.
Nothing here requires a commitment. You choose the model that matches the job, check the per-token rates, and let the bill follow the work. On 59API, a light day stays light and a heavy day is simply metered.
When a card is worth it, and what it changes
Cards are optional on 59API, and they exist for a specific pattern: a stretch of sustained, predictable usage where you would rather work against a fixed allowance than watch a meter. The G-PRO Starter card for GPT costs $83.19 (¥599) for 30 days, with a quota of $2300 and a weekly cap of $880. The C-PRO Starter card for Claude costs $83.19 (¥599) for 30 days, with a quota of $610 and a weekly cap of $230.
If your timeline is shorter, there is a weekly option: the G-PRO $1100 card for 7 days costs $40.56 (¥292), with a quota of $1100 and a weekly cap of $1100. A weekly card is a good match for a focused sprint week and a poor match for a month of scattered requests.
Four rules matter before you buy one. First, a card has its own key and can only call that family of models. Second, a card is paid only from the card: it never falls back to wallet balance, and a balance key never draws on a card. Third, unused quota does not roll over. Fourth, quota and caps are card metering units written in $, not official list-price dollars, so they should not be converted into official prices.
Card usage drains at the card metering price multiplied by 2.4. That is a metering rule, not a comparison with any vendor list price. For a developer whose usage genuinely spikes every day for weeks, a card can make budgeting easier. For someone whose demand comes and goes, pay-as-you-go on the 59API prepaid balance is the simpler fit, because unused balance never expires.
Connect in minutes and start metering
Once you top up, connecting takes a couple of settings. For Codex, OpenAI SDKs, cURL and OpenAI-compatible clients such as Cursor, Cherry Studio, Cline and OpenCode, use base URL https://api.59api.com/v1 and chat endpoint https://api.59api.com/v1/chat/completions. These clients work with GPT, DeepSeek and other non-Claude, non-Gemini models.
Claude models work only through the genuine Claude Code or Claude Desktop client, version 2.1.285 or newer. Set ANTHROPIC_BASE_URL=https://api.59api.com and ANTHROPIC_AUTH_TOKEN=YOUR_59API_KEY. Gemini models use Google's native protocol through Gemini CLI: set GOOGLE_GEMINI_BASE_URL=https://api.59api.com and GEMINI_API_KEY=YOUR_59API_KEY. Gemini models are not OpenAI-compatible and do not work on /v1/chat/completions.
That is the whole model. One account, one API key, one host, and a prepaid balance that is charged per request by real token usage. No monthly fee, and no balance expiry to remember.
If your coding workload genuinely swells and shrinks week to week, a fixed subscription is charging you for a shape of work you do not have. 59API lets the bill match the actual request volume: start with a small top-up, pick the model that fits the task, and upgrade to a card only when sustained usage makes one worthwhile. You can check the current per-token rates and the card options at https://59api.com and decide with real numbers in front of you.